Field conditions for the use of certain empirical bayes estimators

Abstract

Established here are the minimal field conditions for obtaining large sample, normal empirical Bayes limits for mean estimation. For an adverse (uniform) prior distribution, the minimum sample size, the sample number, and relative sample error conditions are established through Monte Carlo results. For as little as two samples of size 25, even where sampling error is half that of the prior, new, easily computed empirical Bayes confidence intervals are shown to improve classical estimation.